Analysis of WBC, RBC, Platelets Using Deep Learning

Abstract

Human blood composition is mainly described into three components which are White Blood Cell (WBCs), Red Blood Cell (RBCs) and platelets. The Complete Blood Cell (CBC)count is used to diagnose the health of a particular person. Proper identification of blood components is the major factor for various uncertainties and health issues in the human body. This paper deals with the analysis of different blood cells using the You Only Look Once (YOLO) framework and has been trained with a dataset of blood smear images taken from BCCD (Blood Cell Count and Detection). Diseases such as dengue, bone marrow disorder, thyroid condition, iron deficiency require blood cell count for the diagnosis. Ordinary methods used in the hospital laboratories require counting of blood cells manually using devices. This led to imprecise outcomes which were strenuous, slow and laborious. The proposed method focuses on obtaining better accuracy with YOLOv5 as compared to previous versions of YOLO models which is based on automatic detection, segmentation and count of each blood cell from blood smear images. Also, Real time implementation can take place and immediately results can be sent for further diagnosis of patient. The main objective of this paper is to identify three major categories of blood cells and improved accuracy is achieved for detection and segmentation of blood cells. The outcome of the experiment on YOLO v5s concludes that highest mAP was observed for 8 batches,75 epochs with mAP value as 93%.